A hybrid genetic algorithm/decision tree approach for coping with unbalanced classes
Déborah Ribeiro Carvalho, Braulio Coelho Avila, Alex Alves Freitas · Kent Academic Repository (University of Kent) · 1999
This paper proposes a new approach for coping wit hthe problem of unbalancaed classes, where some class(es) is(are) much less frequent than other(s). The proposed approach is a hybrid genetic algorithm/decision tree system. The genetic algorithm acts as a wrapper, using the output of a decision tree algorithm (the state-of-the-art C5.0) to compute the fitness of population individuals (candidate solutions to the problem of unbalanced classes). We evaluate the proposed system on a case study application domain about census data.